constraint satisfaction problems - donald bren …welling/teaching/271fall09/csp...constraint...

1

Constraint Satisfaction Problems

Chapter 5 Section 1 – 3

2

Constraint satisfaction problems (CSPs)

  CSP:   state is defined by variables Xi with values from domain Di   goal test is a set of constraints specifying allowable combinations of

values for subsets of variables

  Allows useful general-purpose algorithms with more power than standard search algorithms

3

Example: Map-Coloring

  Variables WA, NT, Q, NSW, V, SA, T

  Domains Di = {red,green,blue}

  Constraints: adjacent regions must have different colors   e.g., WA ≠ NT

4

Example: Map-Coloring

  Solutions are complete and consistent assignments, e.g., WA = red, NT = green,Q = red,NSW = green,V = red,SA = blue,T = green

5

Constraint graph

  Binary CSP: each constraint relates two variables   Constraint graph: nodes are variables, arcs are constraints

6

Varieties of CSPs

  Discrete variables   finite domains:

  n variables, domain size d O(d n) complete assignments   e.g., 3-SAT (NP-complete)

  infinite domains:   integers, strings, etc.   e.g., job scheduling, variables are start/end days for each job: StartJob1 + 5 ≤ StartJob3

  Continuous variables   linear objective & constraints solvable in polynomial time by linear

programming   There are very good, off-the-shelves, methods for convex

optimization problems.

7

Varieties of constraints

  Unary constraints involve a single variable,   e.g., SA ≠ green

  Binary constraints involve pairs of variables,   e.g., SA ≠ WA

  Higher-order constraints involve 3 or more variables,   e.g., SA ≠ WA ≠ NT

8

Example: Cryptarithmetic

  Variables: F T U W R O X1 X2 X3   Domains: {0,1,2,3,4,5,6,7,8,9} {0,1}   Constraints: Alldiff (F,T,U,W,R,O)

  O + O = R + 10 · X1   X1 + W + W = U + 10 · X2   X2 + T + T = O + 10 · X3   X3 = F, T ≠ 0, F ≠ 0

9

Backtracking (Depth-First) search

WA WA WA

WA NT

WA NT

D

D^2

•  Special property of CSPs: They are commutative: This means: the order in which we assign variables does not matter.

•  Search tree: First order variables, then assign them values one-by-one.

WA NT

NT WA

=

WA NT

D^N

10

Backtracking example

11

Backtracking example

12

Backtracking example

13

Backtracking example

14

Improving backtracking efficiency

  General-purpose methods can give huge gains in speed:   Which variable should be assigned next?   In what order should its values be tried?   Can we detect inevitable failure early?

  We’ll discuss heuristics for all these questions in the following.

15

Which variable should be assigned next? minimum remaining values heuristic

  Most constrained variable: choose the variable with the fewest legal values

  a.k.a. minimum remaining values (MRV) heuristic

16

Which variable should be assigned next? degree heuristic

  Tie-breaker among most constrained variables

  Most constraining variable:   choose the variable with the most constraints on

remaining variables (most edges in graph)

17

In what order should its values be tried? least constraining value heuristic

  Given a variable, choose the least constraining value:   the one that rules out the fewest values in the

remaining variables

  Leaves maximal flexibility for a solution.   Combining these heuristics makes 1000

queens feasible

18

Rationale for MRV, DH, LCV

  In all cases we want to enter the most promising branch, but we also want to detect inevitable failure as soon as possible.

  MRV+DH: the variable that is most likely to cause failure in a branch is assigned first. The variable must be assigned at some point, so if it is doomed to fail, we’d better found out soon. E.g X1-X2-X3, values 0,1, neighbors cannot be the same.

  LCV: tries to avoid failure by assigning values that leave maximal flexibility for the remaining variables. We want our search to succeed as soon as possible, so given some ordering, we want to find the successful branch.

19

Can we detect inevitable failure early? forward checking

  Idea:   Keep track of remaining legal values for unassigned variables that are connected to current variable.   Terminate search when any variable has no legal values

20

Forward checking

  Idea:   Keep track of remaining legal values for unassigned variables   Terminate search when any variable has no legal values

21

Forward checking

  Idea:   Keep track of remaining legal values for unassigned variables   Terminate search when any variable has no legal values

22

Forward checking

  Idea:   Keep track of remaining legal values for unassigned variables   Terminate search when any variable has no legal values

23

c a

d

e

b

CONSTRAINT GRAPH

2) Consider the constraint graph on the right. The domain for every variable is [1,2,3,4]. There are 2 unary constraints: - variable “a” cannot take values 3 and 4. - variable “b” cannot take value 4. There are 8 binary constraints stating that variables connected by an edge cannot have the same value.

Find a solution for this CSP by using the following heuristics: minimum value heuristic, degree heuristic, forward checking. Explain each step of your answer.

24

c a

d

e

b

CONSTRAINT GRAPH

2) Consider the constraint graph on the right. The domain for every variable is [1,2,3,4]. There are 2 unary constraints: - variable “a” cannot take values 3 and 4. - variable “b” cannot take value 4. There are 8 binary constraints stating that variables connected by an edge cannot have the same value.

Find a solution for this CSP by using the following heuristics: minimum value heuristic, degree heuristic, forward checking. Explain each step of your answer. MVH a=1 (for example)

FC+MVH b=2 FC+MVH c=3 FC+MVH d=4 FC e=1

25

Constraint propagation

  Forward checking only checks consistency between assigned and non-assigned states. How about constraints

between two unassigned states?

  NT and SA cannot both be blue!   Constraint propagation repeatedly enforces constraints

locally

26

Arc consistency

  Simplest form of propagation makes each arc consistent   X Y is consistent iff

for every value x of X there is some allowed y of Y

constraint propagation propagates arc consistency on the graph.

consistent arc.

27

Arc consistency

  Simplest form of propagation makes each arc consistent   X Y is consistent iff

for every value x of X there is some allowed y

inconsistent arc. remove blue from source consistent arc.

28

Arc consistency

  Simplest form of propagation makes each arc consistent   X Y is consistent iff

for every value x of X there is some allowed y

  If X loses a value, neighbors of X need to be rechecked: i.e. incoming arcs can become inconsistent again (outgoing arcs will stay consistent).

this arc just became inconsistent

29

Arc consistency

  Simplest form of propagation makes each arc consistent   X Y is consistent iff

for every value x of X there is some allowed y

  If X loses a value, neighbors of X need to be rechecked   Arc consistency detects failure earlier than forward checking   Can be run as a preprocessor or after each assignment   Time complexity: O(n2d3)

# arcs

d^2 for checking, each node can be checked d times at most

30

Arc Consistency

  This is a propagation algorithm. It’s like sending messages to neighbors on the graph! How do we schedule these messages?

  Every time a domain changes, all incoming messages need to be re-send. Repeat until convergence no message will change any domains.

  Since we only remove values from domains when they can never be part of a solution, an empty domain means no solution possible at all back out of that branch.

  Forward checking is simply sending messages into a variable that just got its value assigned. First step of arc-consistency.

Constraint Propagation Algorithm

31

•  Maintain all allowed values for each variable. •  At each iteration pick the variable with the fewest remaining values •  For variables with equal nr of remaining values, break ties by checking which variable has the largest nr of constraints with unassigned variables •  After we picked a variable, tentatively assign it to each of the remaining values in turn and run constraint propagation to convergence. (This involves iteratively making all arcs consistent that flow into domains that just have been changed, beginning with the neighbors of the variable you just assigned a value to and iterating until no more changes occur.) •  Among all checked values, pick the one that removed the least values from other domains using constraint propagation. •  Now run constraint propagation once more (or recall it from memory) for the assigned value and remove the the values from the domains of the other variables. •  When domains get empty, back out of that branch. •  Iterate until a solution has been found. •  (as an alternative you only do constraint propagation after an assignment to prune domains of other variables but avoid doing it for all values. Use simply forward checking with the LCV heuristic to pick a value)

32

Try it yourself

[R]

Use all heuristics including arc-propagation to solve this problem.

[R,B,G] [R,B,G]

[R,B,G] [R,B,G]

33

34

This removes any inconsistent values from Parent(Xj), it applies arc-consistency moving backwards.

B R G

B G

B R G

R G B

B G R R G B

Note: After the backward pass, there is guaranteed to be a legal choice for a child node for any of its leftover values.

a priori constrained nodes

35

36

Junction Tree Decompositions

37

Local search for CSPs

  Note: The path to the solution is unimportant, so we can apply local search!

  To apply to CSPs:   allow states with unsatisfied constraints   operators reassign variable values

  Variable selection: randomly select any conflicted variable

  Value selection by min-conflicts heuristic:   choose value that violates the fewest constraints   i.e., hill-climb with h(n) = total number of violated constraints

38

Example: 4-Queens

  States: 4 queens in 4 columns (44 = 256 states)   Actions: move queen in column   Goal test: no attacks   Evaluation: h(n) = number of attacks

39

Hard satisfiability problems

  A,B,C,D,E can take value (true, false).   ¬A=true means that A must be false.   (B ∨ ¬A ∨ ¬C) =true means that B=true or A=false or C=false   Consider random conjunctions of constraints:

(¬D ∨ ¬B ∨ C)=true ∧ (B ∨ ¬A ∨ ¬C)=true ∧ (¬C ∨ ¬B ∨ E)=true ∧ (E ∨ ¬D ∨ B)=true ∧ (B ∨ E ∨ ¬C)=true   We want to find assignments that make all constraints true

m = number of clauses (5) n = number of symbols (5)

  Hard problems seem to cluster near m/n = 4.3 (critical point)

Implementing algorithms for random 3-SAT problems will be your project

Hard satisfiability problems

Hard satisfiability problems

  Median runtime for 100 satisfiable random 3-CNF sentences, n = 50

43

Summary

  CSPs are a special kind of search problem:   states defined by values of a fixed set of variables   goal test defined by constraints on variable values

  Backtracking = depth-first search with one variable assigned per level.

  Variable ordering and value selection heuristics help significantly

  Forward checking prevents assignments that guarantee later failure

  Constraint propagation (e.g., arc consistency) does additional work to constrain values and detect inconsistencies

  Iterative min-conflicts is usually effective in practice